Study design
MOSAIC, a health economic modelling collective established to respond to the need for prompt policy guidance for the South African response to COVID-19, carried out this cost-utility/effectiveness analysis of ICU care. The study was conducted using the principles of the International Decision Support Initiative Reference Case for economic evaluation.[6] It considers two generalised strategies for the inpatient management of severe and critical COVID-19 patients: (1) general ward and ICU management (GW+ICU): admitted patients are managed in a combination of general wards and ICU, with escalation to ICU based on established clinical criteria for severity of disease; (2) General ward management: admitted patients are managed in general wards only. Costs are expressed as the cost per case admitted from the health systems perspective, in 2020 South African Rand (ZAR), including both public and private sector costs. Outcomes are expressed as disability adjusted life years (DALYs) and deaths. While the simple measures of “deaths avoided” or “lives saved” are useful and easy to interpret, they miss important treatment effects, such as improvements in morbidity, and they cannot be used to make comprehensive assessments of value for money compared to other treatment options in the health sector. In contrast, representing the impact of interventions as disability-adjusted life years (DALYs) averted allows consideration of health gains due to a reduction of both fatal and non-fatal disease burden; one DALY can be thought of as one lost year of healthy life. Incremental cost-effectiveness ratios (ICERs) are calculated as the difference in costs divided by the difference in health benefits of the treatment strategies, and are compared to a cost effectiveness threshold (CET) derived from an estimate of the marginal productivity of the public health system in South Africa.[7] If the ICER is lower than the defined CET then the marginal opportunity cost of the treatment strategy (in terms of lost health) is expected to be lower than the health benefits of the treatment strategy, indicating that the treatment strategy is likely to represent a cost-effective use of limited resources.[6] The time horizon for the analysis was from admission to discharge or death; while estimates of ongoing morbidity post discharge were included within DALYs, no costs after discharge were estimated. The years of life lost (YLL) from Covid-19 mortality was informed by a secondary actuarial analysis and was not discounted.
Decision analytic model
A Markov modelling framework was implemented in TreeAge Pro 2020 (TreeAge Software, Inc, Williams- town, Massachusetts, USA) and exported to Microsoft Excel for ease of stakeholder engagement and review, as depicted in Figure 1.
The model, available at https://doi.org/10.25375/uct.12382706, runs for a single cycle; which is appropriate to the available secondary data (where estimates are frequently available as rates) as well as the relatively short duration of COVID-19 inpatient care. Patients are ‘randomized’ within the model to each treatment strategy (GW+ICU versus GW); and on admission (to public or private hospitals), patients are modelled as severe or critical. Depending on these factors, patients incur admission costs and accumulate health outcomes as they transition to one of two absorbing states: recover or die. Recovered patients receive a morbidity loss over the duration of their disease and thereafter are assumed to return to their pre-COVID-19 health state while morbidity as well as YLLs are captured for those dying. Further details of these costs and outcomes are provided within Table 1.
Table 1: Costs and outcomes in each Markov state
Markov state
|
Costs per Markov state
|
Transition outcome to recovered
|
Transition outcome to dead
|
Severe patients
|
Public / private sector cost per hospitalisation in general ward for severe patients
|
Disability weight for severe patients applied over duration of 1.5 months
|
Disability weight for severe patients applied over duration of 0.5 months; Years of Life Lost
|
Critical patients
|
Public / private sector cost per hospitalisation in general ward and ICU for critical patients (GW+ICU model) or general ward only (GW model)
|
Disability weight for critical patients applied over duration of 2 months
|
Disability weight for severe patients applied over duration of 0.5 months; Years of Life Lost
|
Model variables
The models rest on 4 different types of variables: mortality rates based on severity of illness (i.e. severe versus critical) and approach towards disease management (i.e. GW+ICU versus GW); utilisation data including proportions of hospitalized individuals that are critical versus severe, proportion managed in public versus private hospitals and length of stay data for each patient type and management approach; unit costs per inpatient day in general wards and intensive care units specific to public and private hospitals; and DALY data including YLL, years lived with disability (YLD) and disability weights (DWs). A brief overview of each type of data is provided below. Evidence relating to disease progression and effectiveness of interventions has rapidly changed over the course of the COVID-19 pandemic. The parameters used in the model represent best available evidence as at June 2020.
(1) Mortality rates
Mortality rates were extracted from the literature. A systematic search for articles published in English between 01/01/2019 and 30/05/2020 in Medline/PubMed was completed using the terms: "COVID-19" OR “novel coronavirus” OR "SARS-COV-2" OR COVID-19 OR 2019-COV OR “2019 novel coronavirus” AND "clinical characteristics" OR "clinical features" OR "clinical outcomes" AND "death" OR "mortality". Additional relevant articles were sourced through a manual search of bibliographies of included articles.
As outlined in Figure 2, the results of the initial search were screened by title and abstract. The full texts of potentially relevant articles were retrieved and assessed for inclusion. When articles reported information from the same study sites but at two different time periods, only the articles with the updated statistics were included in this analysis. A total of sixteen observational studies (cross-sectional or cohort) and case series that reported the outcomes of hospitalized COVID-19 patients were included within quantitative synthesis. Average weighted estimates of the case fatality rate among ICU patients and non-ICU patients/patients dying in general ward were calculated using the formula :![](data:image/png;base64,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)
(2) Utilisation
Utilisation includes the proportions of hospitalized individuals that are critical versus severe and length of stay data for each patient type by type of management (utilization of ICU days by critical patients, utilization of general ward days for severe patients, and for critical patients before/after ICU). These variables were extracted from seven articles[8-16] identified in the above-mentioned systematic search. Average weighted estimates for each variable were calculated. Finally, the proportion of patients using public versus private hospitals was based on the proportion of South Africans with medical scheme membership.[17]
(3) Unit costs
The model considers the costs of inpatient care in public and private hospitals through the inclusion of unit costs per general ward day and per ICU day. These are multiplied against the abovementioned length of stay estimates to generate a cost per admission. Private sector unit costs are based on the tariff rates in the “Guidelines on Public Private Collaboration in Response to COVID-19” published by the Department of Health.[18] Public sector unit costs were calculated using the Health Systems Trust District Health Barometer (12th Edition – 2016/17) datafile[19] which provides hospital-level estimates of expenditure per patient day equivalent (PDE) for all categories of public sector hospitals. These costs were inflated to 2020 prices using the Consumer Price Index[20] and a weighted average unit cost was calculated through weighting unit costs by the percentage of useable beds across levels of care. Because the HST-DHB data do not provide an estimate of the unit cost per ICU day, we estimated this by inflating the average weighted cost by the cost differential between ICU and general ward tariffs in the private sector.
(4) DALYs
DALYs are calculated through the summation of YLL and YLD. YLL were informed by a South African actuarial analysis that utilised age- and co-morbidity adjusted mortality rates observed internationally and applied these to the South African population.[21] This resulted in an average estimate of 5.4 YLL per death due to the relatively younger population in South Africa. A wide range of this parameter was tested in sensitivity analysis to reflect the relative uncertainty associated with transferring international mortality data to the South African context. Duration of morbidity is currently unknown for COVID-19; assumptions were therefore made for these parameters. Disability weights for severe/critical COVID-19 patients were based on relevant estimates for similar conditions from the 2017 Global Burden of Disease study.[22]
Sensitivity analysis
Simple sensitivity analyses were run across all variables to assess the impact of changes on the ICER. Where possible, ranges for sensitivity analysis were based on upper and lower confidence intervals, high or low values or interquartile ranges found within the systematic literature review. For the remaining variables, a 50% increase/decrease was implemented, except for where this would move the variable out of feasible range (e.g. mortality rates can only fall within the range 0-1). Thereafter, threshold analyses were run to estimate the percentage change in variables that would render ICU cost-effective, using the published South African cost-effectiveness threshold (CET)[7] as the cut-off for this determination. Finally, an additional scenario was modelled in order to incorporate the effect of administration of the steroid dexamethasone. This analysis entailed the inclusion of the cost of a course of dexamethasone (ZAR 160.85 for twenty 4 mg vials as per 2020 Essential Medicines List price), as well as rate ratio reductions in deaths from ICU (0.65) or from general wards (0.80) as provided in estimates from a UK based randomized controlled trial.[23]
Ethical considerations:This is a modelled cost-effectiveness/utility analysis using published secondary data; no ethical approval was therefore required.